commutators modulo the center in a properly … · these special cases any element c is a...
TRANSCRIPT
transactions of theamerican mathematical societyVolume 150, July 1970
COMMUTATORS MODULO THE CENTER IN A
PROPERLY INFINITE VON NEUMANN ALGEBRAR)
BY
HERBERT HALPERN
1. Introduction. An element C in a von Neumann algebra stf is said to be a
commutator in stf if there are elements A and B in stf such that C=AB — BA.
For finite homogeneous discrete algebras and for properly infinite factor algebras
the set of commutators has been completely described [l]-[5], [10]. In each of
these special cases any element C is a commutator modulo a central element
depending on C. In this paper we show that given any element C in a properly
infinite von Neumann algebra stf there is an element C0 in the center of stf de-
pending on C such that C— C0 is a commutator in stf. The element C0 is an arbitrary
element in the intersection criTc of the center with the uniform closure of the convex
hull of {U*CU | U unitary in ¿tf} [6, III, §5]. We then present a few facts about
those elements C such that 0 e Xc or what is the same as far as determining com-
mutators is concerned about those elements C such that OeJf^-ics for some
invertible S in stf.
2. Commutators. Let stf be a C*-algebra with identity and let / be a closed
two-sided ideal in stf. The image of the element A e stf in the factor algebra stf(T)
= stf/1 under the canonical homomorphism of stf onto stf/I will be denoted by
A(T). If £ is a maximal ideal of the center of stf, the smallest closed two-sided ideal
in stf containing £ is denoted by [{]. For simplicity we write A([Q) as A(t). The set
of maximal (respectively, primitive) ideals of stf with the hull-kernel topology is
called the strong structure space (respectively, structure space) of stf. If stf is a von
Neumann algebra, then the strong structure space M(stf) of stf is homeomorphic
with the spectrum of the center S of stf under the map M —> M C\ J? [13]. This
means M(stf) is extremely disconnected.
Proposition 1. Let stf be a properly infinite von Neumann algebra and let A be a
fixed element of stf. The function M -> ||^(M)|| of the strong structure space M (stf)
of stf into the real numbers is continuous.
Proof. For every «¿Owe know that the set X={M e M (stf) \ \\A(M)\\ ^«} is
closed. If /=p| X, then \\A(I)\\^a [8, Lemma 1.9] and so M(M)|| ^a for every
MeM(stf) containing /. Thus X={M e M (stf) | I<=M}.
Received by the editors September 27, 1968.
O The author was supported by the National Science Foundation.
Copyright © 1970, American Mathematical Society
55
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
56 HERBERT HALPERN [July
Conversely, let a > 0 ; we show that
Y = {Me M (sí) I \\A(M)\\ ̂ a}
is closed in M(sé). Let / be the strong radical of sé and let 0* be the structure
space of sé(J). The set
Y' = {Ke0>\ \A(J)(K)\ ä a}
is compact (but not necessarily closed) in 0 [16, 4.9.18]. If 0s' is the structure
space of sé, then M -> M(J) defines a homeomorphism of {M e 0' \ M^>J} — h(J)
onto 0* [16, 2.6.6]. But if M e 0*', then the intersection of M with the center of sé
is a maximal ideal. So M g «( J) implies M is of the form /+ [£] for some maximal
ideal i of the center. It is then clear that h(J) is the set of maximal ideals of sé
[10, Proposition 2.3]. Furthermore, the topology of h(J) and M(sé) coincide. This
proves that Y is compact in M(sé) since it is the inverse image of Y' under the
homeomorphism M -> M(J) of M(sé) onto 0. Because M (sé) is homeomorphic
to the spectrum of the center which is HausdorfT, every compact set of M(sé) is
closed. Thus Y is a closed subset of M (jé). Q.E.D.
Remark. If sé is not properly infinite, Proposition 1 is certainly not true.
Let H be a Hilbert space and let A be a bounded linear operator on H. Let F
be a projection on H. Define the numerical gauge r¡A(F) to be
Va(F) = lub {\\Ax — (Ax, x)x\\ | x is a unit vector in F(H)}.
Let ifA(F) be the closure of the convex set
{L4;c, x)\x a unit vector in F(H)}.
For every ce g ifA(F) we have that
(1) IM-«)/! Ú 65VA(F).
This can be obtained by a simple reworking of Lemma 2.3 [2].
Let sé be a properly infinite von Neumann algebra with no <r-finite type III
direct summands; then for each projection F in sé and each element A m sé
define vA(F) to be
(2) vA(F) = lub {||/LE'-/i',4/i'||£ g (/), E ^ F}
where (/) is the set of projections in the strong radical J oí sé. For every irreducible
representation <j> of sé on a Hilbert space such that </>(J) # (0) we have that
(3) WWO) ^ "a(*0
[10, Proposition 3.1].- Define v(A) to be
v(A) = g\b{vA(F)\l-Fe(J)}.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1970] COMMUTATORS IN A VON NEUMANN ALGEBRA 57
Let stf now be the product of a-finite type III algebras ; let
v(A) = lub {\\AE-EAE\\ | E a projection in stf}
for each A e stf. If A is in the complement in stf of the set of all elements of stf
equal to scalar (zero included) multiples of the identity modulo some maximal
ideal of stf, then there is a v > 0 such that v(AP) ^ v for every nonzero central pro-
jection P since there is a projection Ein stf with E~ 1 — E~ 1 such that EA*(\ —E)AE
ï; a/s for some scalar a > 0 [10, Theorem 3.7]. Also it is easy to see from Proposition
3.1 [10] that
(4) Wl) á v(Ä)
for every irreducible representation of stf.
Lemma 2. Let stf be a properly infinite von Neumann algebra such that
(i) either stf has no a-finite type III direct summands or
(ii) stf is a product of a-finite type III algebras.
Let {Pn} be a set of mutually orthogonal central projections of stf of sum P. Then
v(AP) = lub„ v(APn) for every Aestf.
Proof. Suppose stf satisfies condition (i). Let (/) be the set of projections in the
strong radical of stf. Then given e > 0 there is a projection F with 1 — Fe (J) such that
v(AP) + e ^ lub{||(l-£)APF|| | E£ F, Ee(J)}.
But
\\(l-E)AEPn\\ Í \\(l-E)APE\\
for every Pn. So v(APn) g v(AP) + e for every «. Since e > 0 is arbitrary, we have that
lub v(APn) á v(AP). Conversely, given that e > 0 there is for each Pn a projection
Fn with 1 -Fn e (J) such that
lub{\\(l-E)APnE\\ \E^Fn, Ee(J)} ^ v(APn) + e.
Setting F=^FnPn, we have that P—Fe(J) [10, Corollary, Proposition 2.2] and
that
\\(l-E)APE\\ - lub„ \\(l-E)APnE\\ ¿ \ubv(APn) + e
for every E in (/) with E^F. Thus v(AP)-¿bib v(APn) + e. Since e>0 is arbitrary,
we have that v(AP) á lub v(APn). This completes the proof for case (i).
The proof for case (ii) is similar.
Lemma 3. Let stf be a properly infinite von Neumann algebra with no a-finite type
III direct summands. Let G be a projection in stf such that the orthogonal complement
l—G of G is in the set of projections (J) of the strong radical of stf. Let (J') be the
strong radical of the reduced algebra GstfG. For any element A in stf we have that
v'(GAG) = glb{lub{||(l-£)G^GF|| \ E Ú F, Ee(J')} | G-Fe(J')}
is equal to v(A).
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
58 HERBERT HALPERN [July
Proof. First we show that (J') = {Ee(J) | E^G}. Suppose E^(J') for some
E e (J) majorized by G. There is a nonzero projection P in the center of GséG
such that PE~P [10, §2]. But P=QG for some nonzero Q in the center of sé
[6, I, §2, Corollary, Proposition 2]. Then QE~ QG~ Q since G~ 1 [cf. 10, §2].
So E$(J). This is a contradiction. Hence {Ee (J) | E^G}<=(J'). Conversely,
suppose E e (J'). If Q is a central projection in sé with EQ~ Q, then EQ~ Q~ QG.
This implies 0 = 0. So (J')^{Ee(J) \ E<,G}.
Now let e > 0 be given. There is by relation (2) a projection F in sé with 1 — Z7 g (7)
such that vA(F) á v(^) + e. Let G' be the least upper bound of 1 — G and 1 — F. Then
G'g (7) and 1-G'^Fand 1 -G'úG. We see that
G-(l-G') = G'-(\-G)e(J')
by the first paragraph. So
v'(GAG) ̂ lub{||(l-£)G/ÍG/i'|| | Ee(J'), E ^ 1-G'}
= lub{\\G(l-E)AE\\\Ee(J),E^ 1-G'}
á lub {|| (l-E)AE || | Ee(J),E ^ F} á v(^) + «-
Since e>0 is arbitrary, we have that v'(GAG)^v(A).
Conversely, let e>0 be given; there is a projection Fe GséG with G-Fe(J')
and
\\xb{\\(G-E)AE\\ \E£ F, Ee(J')} ^ v'(GAG) + e.
The domain support G' of (1 — G)A is a projection in (7) since G'«<1 — G [6, III,
§1, Proposition 2]; similarly, the domain support G" of G'Fis a projection in (/)
majorized by F. So 1 -(F-G") = (\ -G) + (G-F) + G" is a projection in (7). Then
v(A) á lub{||(l-£)yá£:|| I E ^ F-G", Ee(J)}.
But
GAGE = AE-O-GKE = ^£-(1 -G)AG'FE = AE-(\-G)AG'FG"E = AE.
So we see that
v(A) S lub{||(l-£)GylG7i|| | E <> F-G", Ee(J')}
á lubíKl-^G^G/ill \E-¿ F, Ee(J')} ^ v'(GAG) + e.
Since e > 0 is arbitrary, we have that v(A) á v'(GAG). Q.E.D.
Theorem 4. Let C be an element in a properly infinite von Neumann algebra sé
and let C0 be an element in the intersection Jfc of the center of sé with the uniform
closure of the convex hull of{U*CU \ U unitary in sé}. Then C— C0 is a commutator
in sé.
Proof. We first make a preliminary reduction. There is a maximal central
projection P in si such that (C — C0)P is in the strong radical / of sé [ 10, Corollary,
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1970] COMMUTATORS IN A VON NEUMANN ALGEBRA 59
Proposition 2.2]. However, the structure of the strong radical of stf allows us to
conclude that (C-C0)P is in the strong radical of stfP [10, §2]. Since (C-C0)P
is a commutator in stfP, it is necessary to prove that (C— C0)(l —P) is a commutator
in stf(\ —P). It is easy to see that the uniform closure of the convex hull of
{U*(C— C0)U I U unitary in stf(\ — P)} contains 0. So without loss of generality
we may assume that C is an element in stf such that 0 e Jfc and such that CP e J
for some central projection P in stf implies that P = 0.
Now there is a central projection Q in stf such that
(i) stfQ has no a-finite type III direct summands and
(ii) stf(\ — Q) is the product of ¡r-finite type III algebras.
It is clearly necessary to prove only that CQ and C(l — Q) are commutators in
stfQ and stf(\ — Q) respectively. Here 0 e CfCQ and 0 e $Tca _ 0) is also true when these
sets are formed relative to stfQ and stf(l-Q) respectively. In the ensuing paragraphs
we shall assume that either stf satisfies condition (i) or condition (ii).
Let ICI = a. We construct by induction a sequence {Pn} of mutually orthogonal
central projections of sum 1 such that for each nonzero central projection P
majorized by Pn the number v(CP) lies in the real interval [2~n<x, 2~" + 1a]. The
induction hypothesis may be stated as follows: let Po = 0; then Pn is the largest
central projection majorized by 1 — 2{Pfc | O^A:^«—1} such that v(CP)eIn for
every nonzero central projection P majorized by Pn. Suppose we have constructed
Pi, P2,..., Pn. We find P„ +1. We may assume R = 1 — 2 {Pk I 0 è k S «} is nonzero.
There is a maximal ideal M in the strong structure space M(stf) of stf such that
CR(M) is not in the center of stf(M). Indeed, suppose CR(M) is in the center of
stf(M) for every Me M (stf). Given e>0 there are unitary operators Ult U2,...,
Um in stf and positive scalars au a2,..., am of sum 1 such that ||2 «i^*C<7(|| <e
since 0 e $TC. Thus
\\CR(M)\\ = \\(2<XiUi*CUi)(M)\\ < e
for every Me M (stf). Because e>0 is arbitrary, we have that CR(M) = 0 for every
Me M(stf). This means that CR is in the strong radical of stf. This is contrary to
the choice of C. Hence, we must conclude that CR(M) is not in the center of
stf(M) for at least one M in M(stf). Then there is a projection E in stf such that
\(l-E)CRE(M)\\jtO. By the continuity of M' -+ ||(l-.E)Ciî£(Af')|| on M(stf)
(Proposition 1), there is an open and closed neighborhood I of M in M(stf)
such that for every M' e X the element CR(M') is not in the center of stf(M').
Let Q be the nonzero central projection of stf which determines X by the relation
X={M' e M (stf) | Q i M'} [13]. Then Q is majorized by R and CQ(M') is not
in the center of stfQ(M') for every M' in the strong structure space M(stfQ) of
the algebra stfQ. The latter is true because M' -> M'Q defines a homeomorphism
of
{M' eM(stf) \QfM'} onto M(stfQ) [16, Theorem 2.6.6].
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
60 HERBERT HALPERN [July
Then by 3.1 and 3.7 [10] there is a v>0 such that v(CQ')^v for every nonzero
central projection Q' majorized by Q. It is clearly immaterial whether v(CQ') is
evaluated in séQ or in sé. Let m be the smallest integer for which there is a nonzero
central projection Q majorized by R such that v(CQ')^2~ma for every nonzero
central projection Q' majorized by Q. We then have that v(CQ') e Im for every such
Q'. In fact by the choice of m the projection Q is easily seen to be the least upper
bound of a set {g,} of nonzero mutually orthogonal central projections which
satisfy 2~ma^v(CQi)<2'm + 1a. By Lemma 2 we may conclude that v(CQ)^
2~m + 1a. So for every nonzero central projection Q' majorized by Q we have that
V(CQ') á v(CQ) ^ 2 - m + !«. Now if « +1 < m set Pn +1 equal to zero. If m S « +1, then
m = n+\. Indeed suppose that m<«+l; the projection Pm+Q¥=Pm, and for any
nonzero central projection Q' majorized by Pm+ Q we have that
v(CQ') = lub {v(CPmQ'), u(CQQ')} e Im.
This contradicts the definition of Pm. Therefore m = « + 1. Now we argue as follows.
Let {on} be a maximal set of nonzero mutually orthogonal central projections
majorized by R such that v(CQ') e In + 1 for every nonzero central projection Q'
majorized by some g¡. Let /,n + i = 2 6¡- It is c'ear tnat Pn + i = R- Because v(CQ)
= lub¡ "(CQQi) for any nonzero central projection Q majorized by Pn + 1 (Lemma 2)
and since at least one projection QQt is nonzero, we have that v(C0^2"<n + 1)a.
On the other hand Q ¿ R and so by the induction hypothesis there is a set {/?,} of
nonzero mutually orthogonal central projections of sum Q such that v(CRt) <2~na
for each /?,. Thus v(CQ)^2~na (Lemma 2). This proves that v(CQ) e In + 1. It is
clear that Pn + 1 is the largest central projection majorized by R such that v(CQ)
e In + 1 for every nonzero central projection majorized by Pn + 1.
Suppose that the sequence {Pn} with the required properties has been constructed
by induction. We show that J,Pn = l- If R=l — 'ZPn, then for each «=1, 2,...
we may conclude that v(CR) g 2 " na by performing the construction of the previous
paragraph. This means that v(CR) = 0. The results of the previous paragraph show
that R = 0 by our choice of C. Hence 2 £„= 1.
Let Z be the spectrum of the center of sé and let Xn = {t eZ \ Pn $ Q for
«= 1, 2,.... Now suppose that sé has no a-finite type III direct summands. For
each «=1,2,... there is a projection Fn in séPn with Pn — FneJ such that vCPj¡(Fn)
^2v(CPn) by definition. For each £gZ there is an irreducible representation
fa = ¡/j of sé on a Hilbert space whose kernel is the smallest closed two-sided ideal
[£] of sé which contains £ [11, Theorem 4.7]. Then </> does not annihilate 7. Indeed
/contains a projection E of central support 1, cf. [10, §2]. The two-valued continu-
ous function £' -*■ ||£(£')ll P, Lemma 10] on Z assumes the value 0 on an open
and closed set given by {£' g Z | R $ £'} where R is a central projection. Then
1 (£(1-R)~E)(01|=0 for every i'eZ. Since HUH I £'eZ} = (0) cf. [9, §4,remarks preceding Lemma 9], we have that E(\ —R) = E. But this means that
R = 0 since the central support of £ is 1. This proves that £(£') # 0 for every t,' eZ
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1970] COMMUTATORS IN A VON NEUMANN ALGEBRA 61
and in particular E(l)^0. So </< does not annihilate J. Thus relations (1) and (3)
imply that
(5) \\(i(CPn)-ß)t(Fn)\\ ^ 65VtncPnMFn)) Ê 65,CPn(Pn) á 130KCP0
whenever ß eiV^CFn)(^(F^) = iV and £ e Xn. We show that OeW. Indeed, given
e > 0, there is a set Uu U2,..., Um of unitary elements in stf and positive scalars
au a2,..., am of sum 1 such that 112 aiU*CUi\\ < e. If G¡ is the range projection of
U*(Pn — Fn) for /=1, 2,..., m, then G¡ is a projection in / and G = lub G¡ is a
projection in J. Thus the projection Pn — G is equivalent to Pn. Let x be a unit
vector in the subspace determined by i/>(Pn —G). For each U¡ we have that <l>(U^x
=y¡ is in the orthogonal complement of the subspace determined by t/i(Pn — Fn)
and thus >/>(Fn)yi=yi. Therefore
2 <<KCPùy» yd e *r
since W is convex. But we have that
\2(^iCPn)yi, yt)\ Û 1011 |2«iiW!| NI2 < ••
Because W is closed and because e > 0 is arbitrary, we see that 0 6 IT. The relation
(5) now becomes
||CPnFn(Q|| = \\CFn(Q\\ = \\UCPn)UFn)\\ Ú 130,(CPn),
for every { e Xn. The orthogonal complement of the projection F= 2 Fn is in J
since P„(l —F)eJ for every «=1,2,... [10, Proposition 2.2]. For every nonzero
central projection P majorized by Pn we have that
||CPP|| = lub{||(CF(0|| \PÍi,ieZ}ú \30v(CPn) á 260KCP).
So for any central projection P we have that
||CTP|| = lub {|| CFPPn || |«= 1,2,...}
(6) ^ lub {260v(CPPn) | « = 1,2,...}
S 260KCP)
by Lemma 2.
Now for an algebra which is the product of cr-finite type III algebras, we may
show that \\CP\\^260v(CP) for any central projection P. The proof is entirely
similar to that just given except that relation (4) replaces relations (1) and (3).
Now let us suppose that stf has no a-finite type III direct summands. Let D = FCF
and let 38 be the von Neumann algebra FstfF. By setting Qn = PnF we obtain a
sequence {Qn} of mutually orthogonal central projections in 38 of sum F such that
v(DQ)eIn for any nonzero central projection Q in 38 majorized by Qn. Here
v(DQ) is evaluated in 38 and Lemma 3 is employed. By relation (6) we see that
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
62 HERBERT HALPERN [July
{||2nZ)ßB||} is a bounded sequence and hence B=Jt2nDQn defines an element of
SS such that v(BQ) ^ a for every nonzero central projection Q in 0J. Indeed,
v(BQ) = \ub{v(BQQn)\n= 1,2,...}
= \ub{2»v(DQQn)\n = 1, 2,...} £ a
since at least one projection QQn is nonzero. There is an invertible element S in
3S and a projection G in 3S with F~G~F-G such that U*S~1BSU=0 and
F*S ~1BSV is a commutator in SB. Here Í7 and F are partial isometries in B such
that U*U=V*V=F, UU* = G and KK* = £-G [10, Theorem 3.6]. By multiplying
both U*S'1BSU and K*5,"1/i5K by the central element 2 2"nön we see that
U*S~1DSU=0 and V*S~1 DSV is a commutator in ££. Now let T=S+(\ -F) in
jé; the element T is invertible with inverse T'1 = S~1 + (l—F) where S"1 still
denotes the inverse of S in á?. Let W be a partial isometry in ¿/ with domain support
1 and range support F. Then V1 = Klf is a partial isometric operator of domain
support 1 and range support F-G. Then it is easy to see that V?T~1CTV1
= W*V*S-1DSVW is a commutator in sé. We have that \~F~G<G + (\ -F).
Thus there is a partial isometry Ux in ¿/ with domain support 1 and range support
G + (\-F). Then
u?t~1ctu1 = r/fr-lcr(i-F)j71+^1*(i-ior-1c,n?^1
is an element of the strong radical of sé and therefore, is a commutator in j/ [10,
Theorem 2.5]. We have proved that there is an isomorphism of sé onto the algebra
sé2 of 2x2 matrices over j/ which carries T~XCT into the matrix (Btj) where /ilx
and /J22 are commutators in sé. But this matrix is a commutator in the algebra
sé2. Indeed, let BXi = S11T11-T11S11 and B22 = S22T22-T22S22 for S1U S22, Tllt
T22 in sé. We may assume that S±1 and S22 have disjoint spectra since 5n =
(S'11+j8)7'i1-T11(S'11+ß) for a scalar /3. There is an operator T12 and an operator
T21 in j* such that 511r12-r12522 = /i12 and S22T2l-T21Sxl=B21 [12]. Setting
5,2i = 5'i2 = 0, we find by direct calculation that (S,y)(7'i,)-(riy)(5ii) = (/iy) in sé2.
This proves (Si;) is a commutator in ,s/2 and T'1CT is a commutator in .s/.
Now if jj/ is the product of a-finite type III agebras, the preceding paragraph
allows us to conclude that there is an invertible S in sé such that S~1CS may be
identified with the 2x2 matrix (Btj) over sé with Bxx = 0 and B22 a commutator
in sé. So (Bij) is a commutator in the 2x2 matrices over sé and 5 " 1CS is a commu-
tator in sé. Q.E.D.
3. Elements C with 0 g J^.. The construction of Theorem 4 actually depended
upon choosing a central element C0 corresponding to a given element C in a
properly infinite von Neumann algebra sé such that ES~1(C-C0)SE=0 for some
invertible S in sé and some projection Ein sé equivalent to 1. The next proposition
clarifies this choice.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1970] COMMUTATORS IN A VON NEUMANN ALGEBRA 63
Proposition 5. If C is an element in a properly infinite von Neumann algebra stf
such that ECE=0 for some projection E in stf equivalent to 1, then 0 is an element
of the intersection d~c of the center of stf with the uniform closure of the convex hull
of the set {U*CU \ U unitary in stf}.
Proof. There are projections E' and E" such that E'~E"~E and E' + E" = E
[6, III, §8, Theorem 1, Corollary 2]. Then l~£'á£" + (l -E)^ 1 implies that
£" + (1— E)~l. So there is no loss of generality in supposing that E~l—E~l.
The operator U=E— (1 — E) is unitary in stf and
2^(C+U*CU) = (l-E)C(l-E).
Now let Ex, E2,..., En be orthogonal projections of sum E such that Ex~ ■ ■ ■
~En~E. There are unitary operators Í/, U2,..., Un in stf such that (l-E)Uj
has domain support E¡ for j= 1, 2,..., « since E,~ 1 — E and 1 — E¡~E for j= 1, 2,
...,«. So we have that
\^{n-1Uf(\-E)C(\-E)Ujx\j = 1, 2,. .., «}||2
â«-2iicii22n^ii2^«-2iia2iwi2for every x in the Hilbert space. Thus
[|2«-1tf*(l-£)C(l-£)í/y|| á «_1||C||.
This means that 0 is an element of the uniform closure of the convex hull of
{V* CV\ V unitary in stf} because « is arbitrary. Hence 0 e 0fc. Q.E.D.
Let stf be a properly infinite von Neumann algebra. If we could prove that
0e U {Jfs-^cs I S invertible in stf} for every commutator C in stf, then we would
have a complete characterization of the set commutators. This characterization
is certainly valid for factor algebras. Indeed C is a commutator in the properly
infinite factor algebra stf if and only if stf is not a nonzero scalar multiple of the
identity modulo the unique maximal ideal M of stf. If C e M, then Oe Jfc [10,
Proposition 2.4]. If C is not a scalar multiple of the identity modulo M, then the
canonical form of Brown and Pearcy [2] in conjunction with the preceding prop-
osition shows 0eJfs-ics for some invertible S in stf. The characterization though
is at odds with a conjecture that the set of commutators in stf is the complement
(F') in stf of the set of all elements equal to a nonzero scalar multiple of the identity
modulo some maximal ideal of stf [4]. In fact let {Pn} be a sequence of nonzero
mutually orthogonal central projections of sum 1. (This presupposes that stf has a
sufficiently large center.) Then let En be a projection in stfPn such that En~Pn
~Pn-En («=1, 2,...). Let C=2(«"1Pn + «-2F0. We have that
\J{MeM(stf)\Pn^M}n
is dense in the strong structure space M(stf) of stf by the remarks at the beginning
of §2. Then C(M) = n~1-l(M) + n~2En(M) for every M with Pn i M and clearly
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
64 HERBERT HALPERN [July
C(M) is not a scalar (zero included) multiple of the identity. But S'1CS(M) is
not a scalar multiple of the identity for every M with Pn $ M. If for example
0 g ^-ics, then by the proof of Theorem 4 we would be able to find an invertible
Tand a projection £ equivalent to 1 with ET~1CTE=0. Thus
[n-tEPJl = n-^ET-^TEW S «"Ir"1!! ||£||
for each «= 1, 2,.... This is obviously impossible.
It might be well to remark that there is no canonical matrix form with 0 on the
diagonal in the sense of Brown and Pearcy [2] for operators of class (£').
Lemma 6. Let C be an element in a von Neumann algebra sé. Let DY and D2 be
elements in C#~c and let A be a central element of sé with O^A^ 1. 77ie« ADX +
(l-A)D2eJTc.
Proof. First let A be a projection in the center of sé. There are unitary operators
Ux, U2,..., Un (respectively Vu V2,..., Vm) and positive scalars au a2,..., an
(respectively ßu ß2,..., ßm) of sum 1 such that
12 KiUfCUi - /Al < e (respectively, ¡2 ßtV(*CVt - D2\\ < e).
Here £>0 is a preassigned constant. Then U{=UtA + (l—A) (lûitkn) and
V/= Vi(\ — A) + A (1 úiúm) are unitary in sé with the property
¡Jiatßjv;*u;*cuiv;-(D1A+D2(i-A))l < e.
Since Jfc is closed, we have that AD1 + (l —A)D2 e Jfc.
Suppose that the restriction that A is a projection is removed. Let Z be the
spectrum of the center of sé and let D~(Q denote the Gelfand transform of the
central element D evaluated at £ e Z. Since Jfc is convex, the set
J^ = {/T(0 | D e JQ
is convex and so
A~(0D?(0 + (l-Ar(0D2(Q e ^
for every t,eZ. Thus there is for each e > 0 a finite set Pu P2,..., Pn of orthogonal
central projections of sum 1 and corresponding elements Bu B2,..., Bn in J^
such that
\\(AD1 + (l-A)D2-BJ)Pj\\ <e
for j=l, 2,..., «. Since ^B¡Pje C#~c by the first paragraph and since e>0 is
arbitrary, we see that AD1 + (l—A)D2 is in the closed set Jfc. Q.E.D.
The next proposition corresponds to a result of C. R. Putnam [14]. We use
many of his calculations cf. [15, 1.5.1].
Proposition 7. Let A, B, C be elements in a properly infinite von Neumann
algebrase. If A is seminormal (i.e. if either ±(AA* — A*A) is positive) andif C = AB
— BA, then 0 e ^s-1Csfoi" some invertible S in sé.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1970] COMMUTATORS IN A VON NEUMANN ALGEBRA 65
Proof. There is a projection P in the center of stf such that CP is in the strong
radical of stfP and
{MeM(stf) | C(l -P)(M) ± 0}
is dense in the subset {M e M(stf)\ l—P^M} of the strong structure space
M (stf) of stf. Since Jfc = jfCP + Jfca_P) = 3fca_P) due to the fact that ^P = {0}
[10, Proposition 2.4] and since {Me M (stf) \ \—P$M} is identified with the
strong structure space of stf(\ —P), we may assume that {Me M(stf) \ C(M)=£0}
is dense in M(stf).
As we have previously argued we may assume that either stf has no a-finite
type III direct summands or that stf is the product of a-finite type III algebras.
Now for each nonzero real scalar a and each unitary element U in the center
& of stf lût
F(a, U) = \cc\ll2UA + \a\~ll2B*.
Then
sgn (a)(FF*-F*F) = D(a) + sgn(a)S(U)
where
D(a) = a(AA*-A*A)-a-1(BB*-E?B)
and S(U)=UC+U*C*. The set {\D\ =(D*D)112 \ D e JQ is monotonely de-
creasing inZ. Indeed, let Dx and £>2 be elements of $TC; there is a central projection
P such that |Z>i|P^|Z)2|Pand |£>2|(1 -P)£ \DX\(\ -P). But D1P+D2(\-P)e tfc
and |/)1P+£)2(1-P)| = |/)1|P+|/)2|(1-P) is majorized by both |£>i| and \D2\.
This proves that {\D\ \ D e $fc}'\% monotonely decreasing. Let D0 be the greatest
lower bound of this set [6, Appendix II]. Suppose Dx is a positive central element
which majorizes D(a) for some a. We show that D0£DX. If not, there is an e>0
and a nonzero central projection P such that D0P ä (Dx + e)P. By reducing to stfP
we may assume that P=l. Let Uu U2,..., Un be unitary elements in 2£ and let
au a2,..., an be positive scalars of sum 1 ; then for any central element R in the
sphere of radius 2~1e about 0 we have that
|(2«iC/i)/)1 + /?|2 ^ G>>i£>i+I*|)2 = (D1 + \R\)2 g (D1 + 2-1e)2.
Hence (D*Dx, x)^((D1 + 2~1e)2x, x) for any D in the strong closure JT" of the
convex hull of the set
{UDj. + R \U,Rin2£,U unitary, \R\ S 2"1«}
and for any x in the Hilbert space H of stf. On the other hand we see that (D*Dx, x)
^ ((£>! + s)2x, x) for any D in the strong closure Jf ' of Jf^ and any x in H because
R*R^(D1 + e)2 for every R in X~c. By the standard separation theorem there is a
nonzero strongly continuous functional / on 2£ such that
(7) lub {Re/(/?) | R e ¿f'} á gib {Ref(R) \ R e Jf"}.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
66 HERBERT HALPERN [July
Here Re ß denotes the real part of the complex number ß. Indeed the element 0
is not in the strong closure of Jf — Jf". Since fis also weakly continuous on !%
[6, I, §3, Theorem 1 (i)], there is a unitary U in 2£ and a nonzero vector x in H
such tbat f(R) = (RUx, x) for every Re% [17] and [6, III, §1, Corollary, Theorem
4]. Now let F=F(a, sgn (a)U). We have that
(8) sgn (a)(FF*-F*F) = D(a) + S(U) ¿ Dt + S(U).
Let D be an arbitrary element in ^lgn(axFF'-F-F) = ^. There is an element S in
Jfsw) such that D^Di + S [6, III, §5, Problem 2a]. We may find a T in 3fm-wc
such that
2-l(S+T)eJtuc and 2-1(5,-£) g JTV.C.
[6, III, §5, Problem 2a]. We then have that 2-1i/*(5,+ £) and 2-W*(S-T)* are
elements of Jfc. The latter is true because JTC. = {R* \ ReJfc}. From relation (7)
we obtain that both Re (2~1(S+T)x, x) and Re (2~r(S—T)x, x) are majorized
by gib {Ref(R) \ R e Jf"}. Thus
(Sx, x) = Re (Sx, x) á 2 gib {Re/(£) | £ g jíT"}
since 5 is clearly selfadjoint. But — U*D1 — 2~1eU* is an element of cf". So
(9) 2-1(S;r, x) ^ Re (U(- U*D1-2~1eU*)x, x) = -((/J)1 + 2-1£)x, x).
Therefore,
(Z)*, *) ^ (öiX, x) — e(x, x) — 2(DXX, X) ^ - e(x, x)
by relation (8). Using reasoning similar to that which we used to prove that
{\R\ | ReJTc} is monotonely decreasing, we may prove that Jf is monotonely
increasing. Setting /\0 = lub Jf, we see that (R0x, x)á — e(x, x). We show that this
is impossible by showing (R0x, x)^0. Indeed, in proving this then we may cer-
tainly assume that £ is invertible and that a > 0. Because F is invertible, there is a
unitary operator Vinsé obtained from the polar decomposition of £ [6, Appendix
III] such that V*FF* V= £*£. If £ g $fF.F, then there are unitary operators
U-i, U2,..., Un in sé and positive scalars au a2,..., an of sum 1 such that
12 aiUl*F*FUl-R\\ < •'
for any preassigned constant e' > 0. But this means that
l^atiVUyFF^VUd-RW < e'.
Because e'>0 is arbitrary we have that ReJfFF.. By symmetry it is then clear
that JfFF. = JfF.F. Now for any e >0 there is an element £i in $fFF. such that
(Rxx, x) ^ (R2x, x)-e'(x, x)
where R2 is the least upper bound of the monotonely increasing set CtfFF.. But
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
1970] COMMUTATORS IN A VON NEUMANN ALGEBRA 67
there is an element R e JTF.F such that /?!-/? e Jf [6, III, §5, Problem 2a]. How-
ever we have that
(R0x, x) £ ((/?i- R)x, x) ^ (R2x, x)-(Rx, x)-e'(x, x) ^ -e'(x, x)
since R2^R. Because e'>0 is arbitrary, we see that (R0x, x)^0. This is a contra-
diction. We must conclude that D0SD1.
We now show that Do = 0. We may assume that AA*—A*A^0. Then for a>0
we have that
D(a) á -a~\BB*-B*B) % 2a-1||£||2.
Thus we see that D0¿2a~1\\B\\2 for every a>0. Therefore D0 = 0. If g is a nonzero
central projection in stf and if e>0 there is a net {Qn} of mutually orthogonal
central projections of sum Q such that each set ^con contains an element Dn of
norm not exceeding e. Indeed, if {£?„} is a maximal set of mutually orthogonal
nonzero central projections majorized by Q with this property, then the assumption
that Q'=Q-2 Qn^O gives a contradiction. Since glb{|/>| | DeJfCQ.} = 0, there
isa De JfCQ- such that \D\ ^2~1eQ' is not true. This means that there is a nonzero
central projection Q" majorized by Q' such that \D\Q"-¿eQ". This contradicts
the maximality of {Qn}. Hence we have that 2 Qn= Q-
Now suppose stf has no a-finite type III direct summands. In Theorem 4 we
constructed a sequence {Pn} of mutually orthogonal central projections of sum 1
and a projection F whose orthogonal complement 1 — F was in the strong radical
such that \\(C—R)FP\\^260v(CP) whenever P is a central projection majorized
by Pn and whenever ReXCf. Also either Pn = 0 or v(CP)e [2-"||C||, 2-n + 1||C||]
for every nonzero central projection P majorized by Pn. By the preceding para-
graph there is a set {Pnj} of mutually orthogonal central projections of sum Pn
such that each set ^Cpn! contains an element Dnj of norm not exceeding v(CPn).
Then for each nonzero central projection P majorized by Pn we have that
\\CFP\\ = lub, \\CFPPni\\ è lub ||(C-/)n;.)PPni||+lub \\DnjPPnj\\
S 260v(CP) + v(CPn) ¿ 262v(CP)
by relation (6). By the same reasoning as found in Theorem 4, we may find an
invertible Win stf such that EW~1CWE=0 for some projection Ein stf which is
equivalent to 1. However, this means that 0 6 Jfw~1cw by Proposition 5.
If A is the product of a-finite type III algebras a similar proof holds. Q.E.D.
Corollary. If F is an element in a properly infinite von Neumann algebra stf,
then there is an invertible S in stf such that ■%s~1(f-f-ff')s contains 0.
Proof. If A = 2~1i(F-F*) and B = 2-\F+F*), then 2~H(FF*-FF*) = AB
— BA. Now Proposition 7 applies.
Added in proof (April 25, 1970). I have improved Proposition 7 by showing
that 0 e Jfc.
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use
68 HERBERT HALPERN
Bibliography
1. A. Brown, P. Halmos and C. Pearcy, Commutators of operators on Hilbert space, Canad.
J. Math. 17 (1965), 695-708. MR 34 #3311.
2. A. Brown and C. Pearcy, Structure of commutators of operators, Ann. of Math. (2) 82
(1965), 112-127. MR 31 #2612.
3. -, Commutators in factors of type III, Canad. J. Math. 18 (1966), 1152-1160. MR
34 #1864.
4. A. Brown, C. Pearcy and D. Topping, Commutators and the strong radical, Duke Math. J.
35 (1968), 853-859. MR 38 #1534.
5. D. Deckard and C. Pearcy, On continuous matrix-valued functions on a Stonian space,
Pacific J. Math. 14 (1964), 857-869. MR 30 #2356.
6. J. Dixmier, Les algebres d'opérateurs dans l'espace hilbertien. Cahiers Scientifiques,
fase. 25, Gauthier-Villars, Paris, 1957. MR 20 #1234.
7. -, Les C*-algèbres et leurs représentations, Cahiers Scientifiques, fase. 29, Gauthier-
Villars, Paris, 1964. MR 30 #1404.
8. J. M. G. Fell, The structure of algebras of operator fields, Acta Math. 106 (1961), 233-280.
MR 29 #1547.
9. J. Glimm, A Stone-Weierstrass theorem for C*-algebras, Ann. of Math. (2) 72 (1960),
216-244. MR 22 #7005.
10. H. Halpern, Commutators in properly infinite von Neumann algebras, Trans. Amer. Math.
Soc. 139(1969), 55-73.
11. -, Irreducible homomorphisms of a von Neumann algebra into its center, Trans. Amer.
Math. Soc. 140 (1969), 195-221. MR 39 #3322.
12. G. Lumer and M. Rosenblum, Linear operator equations, Proc. Amer. Math. Soc. 10
(1959), 32-41. MR 21 #2927.13. Y. Misonou, On a weakly central operator algebra, Tôhoku Math. J. (2) 4(1952), 194-202.
MR 14, 566.
14. C. R. Putnam, On commutators of bounded matrices, Amer. J. Math. 73 (1951), 127-131.
MR 12, 836.
15. -, Commutation properties of Hilbert space operators and related topics, Ergebnisse
der Mathematik und ihrer Grenzgebiete, Band 36, Springer-Verlag, New York, 1967. MR 36
#707.16. C. Rickart, General theory of Banach algebras, The University Series in Higher Math.,
Van Nostrand, Princeton, N. J., 1960. MR 22 #5903.
17. S. Sakai, On linear functionals of W*-algebras, Proc. Japan Acad. 34 (1958), 571-574.
MR 21 #5915.
Illinois Institute of Technology,
Chicago, Illinois 60616
License or copyright restrictions may apply to redistribution; see https://www.ams.org/journal-terms-of-use